43 lines
2.6 KiB
Markdown
43 lines
2.6 KiB
Markdown
# Changelog
|
|
|
|
All notable changes to this project will be documented in this file.
|
|
|
|
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
|
|
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
|
|
|
|
## [0.2.1] - 2026-05-22
|
|
|
|
### Added
|
|
- `scripts/install_server.sh` — one-shot idempotent server-side install (ffmpeg check + torchaudio install + smoke test).
|
|
- `README.zh-CN.md` — Chinese quick-start guide (shortest working path).
|
|
- `README.md` rewritten with explicit shell `$` prompts, ⚠️ pre-install warning block, step-by-step `Install (the proven order)` with per-step verification, and a Tested-with compatibility table — matching the ComfyUI-MiMoASR documentation style.
|
|
- `INSTALL_NOTES.md` reorganized as a 9-step debug log with the three MiMo install pitfalls called out (transformers downgrade, flash-attn build, upstream repo clone).
|
|
|
|
### Changed
|
|
- `README.md` now uses 9 distinct verification steps. Each install step ends with a `Verify:` block.
|
|
- Troubleshooting tables now distinguish errors that surface at ComfyUI startup vs at workflow runtime (lazy-imports).
|
|
|
|
## [0.2.0] - 2026-05-21
|
|
|
|
### Changed
|
|
- **BREAKING**: dropped the bundled `FasterWhisperTranscribe` node. ASR is now handled externally by [ComfyUI-MiMoASR](https://github.com/aadebuger/ComfyUI-MiMoASR) (`MiMoASRLoader` + `MiMoASRTranscribe`).
|
|
- **BREAKING**: JSON schema bumped to `2.0`. `segments[]`, per-segment `confidence`, `avg_logprob`, `no_speech_prob`, `compression_ratio`, and word-level timestamps are removed because MiMo's `asr_sft` does not expose them. Top-level `text` field now carries the full transcript.
|
|
- Dropped `faster-whisper` from `requirements.txt`. Added `torchaudio`.
|
|
|
|
### Added
|
|
- `VTT_AudioFromPath` — bridge node: loads an arbitrary WAV/audio file (from absolute path) into ComfyUI's `AUDIO` dict so it can be fed into `MiMoASRTranscribe`. Optional `target_sample_rate` and `force_mono`.
|
|
- `VTT_BuildTranscriptJSON` — pack ASR text + source metadata into the v2.0 JSON schema.
|
|
- Example workflow now chains `VTT_*` nodes with `MiMoASRLoader` + `MiMoASRTranscribe`.
|
|
|
|
### Removed
|
|
- `nodes/whisper_transcribe.py` and all references to faster-whisper.
|
|
|
|
## [0.1.0] - 2026-05-21
|
|
|
|
### Added
|
|
- Four nodes: `VTT_LoadVideo`, `VTT_ExtractAudio`, `VTT_FasterWhisperTranscribe`, `VTT_SaveTranscript`.
|
|
- faster-whisper backend with per-segment confidence and word-level timestamps.
|
|
- Multi-GPU device selection; `_MODEL_CACHE` for warm model instances.
|
|
- ffmpeg-based audio extraction with on-disk caching.
|
|
- CLI client (`cli/video_to_text_cli.py`) using `PreviewAny` to pull outputs over WebSocket.
|